Why You Don't Have to Talk Yourself Into Affirmations

The sentence in the mirror

Morning, in front of the mirror. "I am confident. I am attractive. I am successful." Ten times, better twenty. The internet promises that sooner or later the brain will take the message on board. You just have to keep at it, because the subconscious learns through repetition.

The idea is old and appealing. In the early twentieth century, the French pharmacist Émile Coué recommended saying twenty times every morning and every evening: "Tous les jours, à tous points de vue, je vais de mieux en mieux." Every day, in every way, I am getting better and better. To keep count, he suggested a string with twenty knots. Today there are apps, inspirational calendar sayings and motivational videos with millions of views built on the same idea. Whether it works is rarely asked. The question seems unnecessary: what harm could a kind sentence do?

The Canadian psychologist Joanne Wood set out to test exactly that. Together with colleagues, she asked participants to repeat a single sentence: "I'm a lovable person." Beforehand, the researchers measured participants' self-esteem. Afterwards, they measured how they felt.

The result was unsettling. People with high self-esteem benefited a little. People with low self-esteem did not benefit. On the contrary, their mood was in some respects worse than in the comparison group. The very people the mirror sentence is meant for came out of the experiment worse off.

This finding should not be made bigger than it is. The experimental samples were modest, what was measured was a short-term state, and two experiments do not establish a law. Anyone who says a kind sentence in the morning and feels good doing it need not stop. But the result deserves to be taken seriously as a warning sign, because it contradicts a widespread assumption: that positive sentences help at best and do nothing at worst.

One possible explanation lies in the gap between the positive sentence and the existing self-image. Someone who is convinced they are not lovable already holds memories and judgements that contradict the new claim. The positive sentence does not erase this information. It may even make it more salient.

A second finding by Wood and her colleagues fits this picture. People with low self-esteem did not only feel worse when they repeated the positive sentence. It was also unhelpful to ask them to think exclusively about why the sentence was true. They fared better when they were allowed to entertain both supporting and contradicting thoughts.

So the problem does not seem to be that a positive thought is useless in principle. It becomes a problem when a claim is supposed to override an existing self-image while one's own experience keeps pushing back.

This leaves a question that will accompany the whole article:

Why should a brain adopt a claim for which it is simultaneously holding counterevidence?

Before answering it, it is worth looking at the small word the whole debate turns on. When someone says affirmations "work", they can mean three very different things. A sentence can change how someone evaluates themselves. Remembering what truly matters to you can stabilise your self-image when it comes under pressure. And a verbal prompt can help bring about something in the body or in experience that you can actually feel.

These three kinds of effect have different mechanisms and very different levels of evidence. The following sections take them one at a time. By the end it will become clear that the most interesting of them has little left in common with the mirror sentence.

A detour that works

Research has found a variant that does have an effect. But it works for an entirely different reason.

In the late 1980s, the American social psychologist Claude Steele put forward an idea that at first has little to do with mirror sentences. People, Steele argued, have a basic need to see themselves as broadly intact: as someone who acts sensibly, stands by their values and is fundamentally okay. When this self-image is threatened, for instance by unwelcome news, a mistake or criticism, it defends itself. We play the news down, doubt the source, look for excuses.

Steele's key point: this self-image rests on many pillars. Someone under pressure in one area can steady themselves in another. And someone who has steadied themselves elsewhere has less need to defend the threatened area so stubbornly.

This led to an experimental procedure that has since been used many times. Participants receive a list of values, such as family, creativity, humour, friendship, religion or knowledge. They choose the value most important to them and spend a few minutes writing about why it matters and when it has guided their actions. That is all.

Notice what is missing from this exercise. Nobody tells themselves something they do not believe. Nobody claims to be successful, beautiful or lovable. The person recalls something that is already true. There is no inner contradiction, because nothing is claimed that runs against their own experience.

In an early experiment, women who regularly drank coffee and women who did not read an article written specifically for the study that linked caffeine to breast cancer risk. (No such link actually exists.) Beforehand, half of them had written about a value that was especially important to them, the other half about an unimportant one. The coffee drinkers rejected the message more strongly than the others. But if they had first written about an important value, they were more likely to accept the information and more often reported an intention to drink less coffee. The exercise had changed nothing about the message. It had reduced the need to defend against it.

This pattern runs through the research. A meta-analysis of 144 experiments on health topics found that self-affirmation increases willingness to accept unwelcome information, strengthens intentions to change behaviour and, in some of the studies, also influences actual behaviour. A more recent meta-analysis with around 18,000 participants also found effects on self-perception and well-being.

The numbers deserve a second look. The effects are statistically reliable but small. And they come mostly from studies of people without a mental illness. Anyone who wants to derive a therapy method from them is going one step further than the data allow.

How much the size of such effects depends on circumstances is shown by the best-known study in this field. Geoffrey Cohen and colleagues had school students complete the values exercise at the start of the school year. Among African American students, who faced negative stereotypes about their academic ability in everyday school life, grades subsequently improved and the achievement gap with their white classmates narrowed considerably. The finding was published in Science and widely celebrated.

Later, Paul Hanselman and colleagues repeated the intervention on a large scale, in a school district where an earlier study using the same procedures had found positive results. This time there was no effect. The study was large enough to rule out effects above a small effect size of 0.10. And none of the theoretically expected moderators could explain why the same exercise had worked once in the same place and not the second time.

This does not mean the original finding was wrong. But it also does not mean that every null result may be explained after the fact by some unknown circumstance. Doing so makes a theory impossible to refute. The honest conclusion is this: as a blanket recipe the exercise does not hold up, and exactly when it works is not sufficiently understood.

What remains is still remarkable. Self-affirmation shows that language can influence how people deal with a threat to their self-image without talking them into a new positive belief about themselves. It connects to something that already carries meaning, and can thereby reduce the need to fend off unwelcome information.

This has little in common with the mirror sentence. And it helps only to a limited extent with the question from the first section. Self-affirmation sidesteps the contradiction rather than resolving it. It does not need to create a new experience that corrects an old belief.

For that, we have to go to a place where sentences have long been credited with special power: hypnosis. And there we first have to clear away a legend.

The trance legend

Coué had a clear idea of why his formula should work. It was not to be spoken with emphasis or willpower, but as mechanically as possible. The very monotony was supposed to let the sentence slip past the critical mind and sink into the unconscious. There, Coué believed, it would take effect.

This idea has outlived Coué. It is found today in almost every popular explanation of hypnosis. Trance, the story goes, opens a door. The critical mind steps aside, and whatever is said then goes straight to where beliefs and habits are formed. If this were true, an affirmation in trance should be far more effective than the same affirmation in front of the mirror.

The hypothesis has one great advantage: it can be tested. You give the same suggestions once after a hypnotic induction and once without. If trance is the decisive door, the difference should be enormous.

A distinction that often gets blurred in everyday language helps here. Hypnosis can mean the procedure: an induction, usually involving relaxation and focused attention, followed by suggestions. Trance, by contrast, refers to the state this procedure supposedly produces. Some people also use the word for states of focused absorption that occur in everyday life without any hypnosis at all. Even experts disagree. The hypnosis division of the American Psychological Association described hypnosis in 1993 essentially as a procedure, but in 2015 as a state of consciousness with an enhanced capacity to respond to suggestion. Prominent hypnosis researchers promptly criticised the new version as a step backward, because it presupposes a special state whose existence is precisely what is in dispute. For the studies that follow, the distinction is crucial. They test the procedure, more precisely: what the induction adds to the suggestions. Whether a special state arises along the way is a separate question.

This is exactly the comparison Wayne Braffman and Irving Kirsch made in the late 1990s. They measured the same people's responses to suggestions without hypnosis and after a hypnotic induction. The hypnotic induction increased responsiveness, but only slightly. Reviews put the typical gain at roughly one and a half additional responses on a twelve-item scale, often less. Someone who responds to six of twelve suggestions without an induction might respond to seven with one. How strongly a person already responded without hypnosis predicted a substantial part of their response after the induction. In a not insignificant minority, responsiveness even dropped after the hypnotic induction.

A study by Steven Jay Lynn and colleagues was even clearer. They gave one group the standard hypnotic induction and another group the same suggestions with no induction at all. Leaving out the induction made no appreciable difference. A further group was told beforehand something that matched the popular idea exactly: whether hypnosis succeeds depends on reaching an altered state of consciousness. This group responded to fewer suggestions than the group with the standard induction. The samples were small, and a single finding should not be overstretched. But the direction is striking. Anyone who believes they must first reach a special state faces a hurdle they can miss. And anyone who thinks they are missing it expects less.

This does not discredit hypnosis. It discredits one particular explanation of hypnosis.

Because what remains once trance is removed is a surprising amount. In response to suggestion, people feel an arm becoming heavy, a hand rising as if by itself, a pain losing its edge. These responses can be reported subjectively and in part observed or measured objectively, and in some people they occur reliably. They simply do not require an open door to the unconscious. Modern hypnosis research has several competing explanations and has not yet settled on one. It broadly agrees that expectation, focused attention, imagination and stable individual differences in responsiveness play a major role. Trance as a floodgate is not among the established building blocks.

This is the most important distinction in this article, and it applies well beyond hypnosis: whether something works and why it works are two different questions. Coué may have been right about the first and wrong about the second. A method can work even though the story told about it is not true.

For our starting question, this is good news. If trance is not the decisive ingredient, the effect of suggestions must come from somewhere else. And that somewhere else is more interesting than any floodgate.

When something follows a sentence

Up to this point we have asked whether a sentence is believed. From here on we ask something different: whether a sentence can set off something you experience.

The Berlin neurologist and psychiatrist Johannes Heinrich Schultz developed a method he called "concentrative self-relaxation", known today as autogenic training. Those who learn it quietly repeat formulas such as: "My right arm is very heavy." On the surface, this is an affirmation. A sentence is repeated, and it claims a state.

The difference from the mirror sentence lies in what happens after the sentence. "I am confident" cannot be checked in the moment, and when it can, the check often comes out negative. "My arm is heavy" can be checked immediately. And the check often comes out positive.

There is a simple reason for this. Anyone who sits or lies still, lets the muscles loosen and turns their attention to their arm notices sensations that are already there: the weight of the arm on the surface, the easing of tension, a slight warmth. The formula does not create these signals out of nothing. It predicts something that is likely to happen under these conditions and directs attention precisely there. Then comes a moment every practitioner knows: "It really is getting heavy."

At that moment the sentence has gained something the mirror sentence never gets. It has been confirmed.

Irving Kirsch, whom we have already met in hypnosis research, formulated part of this mechanism theoretically in the mid-1980s. His response expectancy theory focuses on expectations about what will happen next: Will my arm become heavy? Will I relax? Will the pain ease? According to Kirsch, such expectations do not merely describe an upcoming response. They can contribute to bringing it about.

This is especially interesting for responses we do not experience as deliberately produced. Someone who expects relaxation to set in or pain to ease does not have to consciously decide to produce the corresponding sensation. The expected experience can arise and, precisely because of that, give the impression: it is happening.

For autogenic training, a further sequence can be proposed. The formula creates an expectation. It directs attention to fitting sensations. Existing bodily signals are noticed more easily. When expectation and experience match, the formula gains credibility. With repeated practice, the response can become easier and easier to call up, until a single cue is enough. This sequence is a plausible link between existing findings, not an experimentally confirmed process model.

It does, however, have the advantage of making a testable prediction: suggestions for changes that are physically likely and easy to notice should be easier to establish than suggestions for unlikely changes. One finding from hypnosis research is at least consistent with this. On the widely used Harvard Group Scale, so-called ideomotor items, such as an outstretched arm lowering in response to a suggestion of heaviness, are among the easiest. Across 16 national norm samples, on average around 71 percent passed such items, compared with around 34 percent for cognitive items such as hallucinations or amnesia. This is not proof, because the items differ in many respects at once: in the imagination required, in motor involvement, in the criterion for success, and in how far the experience of one's own authorship has to change. But it fits.

That procedures of this kind achieve something clinically is shown by research. A meta-analysis by Stetter and Kupper, with 60 included controlled studies, 35 of them randomised, found autogenic training more effective than comparison conditions across a wide range of complaints, although the studies were very heterogeneous and spanned several decades. A review of self-hypnosis by Eason and Parris covering 22 randomised trials reported medium to large effects overall, including for pain, stress and anxiety. Notably, the procedures tended to do better when people learned and practised self-hypnosis as a skill of their own. That comparison, however, comes from the pattern across studies, not from a direct test.

The effect of suggestion has been measured most precisely for pain. Trevor Thompson and colleagues analysed 85 controlled experiments with more than 3,600 participants, in which healthy people were exposed to experimentally induced pain. Hypnosis reduced pain on all measured dimensions. But the effect was distributed very unevenly. When a direct analgesic suggestion was given, pain fell by around 42 percent in highly responsive people and by around 29 percent in moderately responsive people. In people with low responsiveness there was hardly any benefit.

These numbers say two things. Suggestion can produce effects in some people that one would hardly think possible in everyday life. And it is not a universal force that works the same in everyone as long as they relax deeply enough. Whether these differences carry over to chronic pain in clinical practice is also less clear than the laboratory findings might suggest.

From all this, levels of experiential grounding can be described:

"I am calm" claims a state. Whether the nervous system follows remains open.

"With every out-breath I become calmer" links the desired change to a process that can be observed.

"My right arm is very heavy" predicts a sensation that is likely under the given conditions and makes the confirmation tangible.

A targeted analgesic suggestion, finally, aims at changing an experience that is already present.

This ordering says nothing about the strength of the effect. It describes how closely a sentence is tied to an experience. And the closer that tie, the less it matters whether you believe the sentence.

One last question remains open. When someone experiences pain easing, do they attribute this to themselves, to the hypnotist or to the situation? Whether a single experience turns into lasting change might depend on the answer.

First it happens, then I can do it

Someone who experiences an arm rising under hypnosis rarely describes it with the words "I raised my arm". Far more often they say: "It rose." The movement is experienced as happening by itself. This altered experience of one's own authorship is among the most striking features of hypnotic responses, and it long fed the idea that in hypnosis some outside agency takes the wheel.

The psychologists Zoltán Dienes and Josef Perner have proposed a more sober explanation, which they call "cold control". Put simply: a person can control an action or a mental change themselves without at the same time forming the thought "I am doing this deliberately right now". The control takes place, but awareness of one's own intention is missing. The result then feels like something that happens, not something one does. No special state of consciousness is required.

This theory matters for our question. If Dienes and Perner are right, hypnotic responses can be self-generated without being experienced as self-generated. The person may then possess an ability whose authorship they are not aware of in the moment. Their own system has dampened the pain, raised the hand or produced some other response, while the experience says: it is happening.

This is one explanation among several. Other models of hypnosis place the emphasis elsewhere. For our argument, cold control is interesting because it allows a distinction we need for the next step: a response can be produced by the person themselves without them initially experiencing themselves as its author.

This brings us to one of the most influential models in psychology. Albert Bandura described how the belief that one can cope with a situation draws on several sources: the example of others, encouragement, one's own bodily state. He generally regarded one's own experiences of mastery as the most influential source. Someone who has managed something is more likely to trust themselves to do it again.

Here the circle closes back to the mirror sentence. Repeating "I can regulate my anxiety" gives the brain no evidence. An experience in which a response previously felt to be uncontrollable actually eases does provide evidence.

Probably only under one condition, though. For an experience to work as one's own mastery, the person presumably has to understand it as their own achievement. Someone who believes the hypnotist drove the pain away has experienced something impressive but learned little about themselves. Someone who believes a special trance did it will think they need a trance again next time.

A classic experiment shows how much can hinge on this attribution. Gerald Davison and Stuart Valins had male students take a test of pain threshold and shock tolerance. The students then took a supposed drug, which was in fact a placebo, and repeated the test. Without their knowledge, the intensity of the electric shocks had been halved. So all of them experienced an apparent improvement and attributed it to the drug. Afterwards, half were told they had received a placebo. These men now had to attribute the improvement to themselves. In the next test, they rated the shocks as less painful and tolerated considerably more than those who still believed in the drug. The experiment had nothing to do with hypnosis, and it did not examine any long-term course. But it shows that the same experienced change can have different after-effects depending on who gets the credit.

This allows a sequence to be described, which we explicitly mark as a hypothesis. I am not aware of any series of experiments that has tested it as a whole.

With hypnosis guided by another person: "It is happening." Under self-hypnosis: "I am making it happen, even if it does not feel that way." After repeated practice: "I can influence it."

Self-hypnosis is the most interesting case here. The person started the exercise themselves, gave the suggestion themselves and experienced the change themselves. They may still attribute the success to the therapist who taught them the method, to a recording, or simply to "hypnosis". But self-attribution is especially close at hand here. That the change can nonetheless feel as though it happens by itself would be consistent with cold control.

This sequence yields a prediction that could be tested. Take people for whom a suggestion works equally well. Half of them then receive an interpretation that credits their own regulation with the success. The other half are told it was due to the hypnotist or a special state. Four things would be measured: immediate self-efficacy, a renewed response without the hypnotist, transfer to a similar new task, and the outcome after a few weeks. Only then could one tell whether the attribution merely changes how people judge themselves or actually changes their ability to regulate themselves. If both groups do equally well, the model described here loses a substantial part of its explanatory power.

Such a study would be the empirical test of this article's central thesis. And it shifts attention to something almost always overlooked in the debate about affirmations: it is not the sentence that decides, and not even the experience alone. What one learns about oneself from the experience may matter too.

What this means in practice

At the end of such a journey comes the question of what to do differently tomorrow. The answer begins with a cautionary tale that comes not from affirmation research but from hypnosis research.

In the mid-1990s, Irving Kirsch and colleagues published a meta-analysis of 18 studies. They compared cognitive-behavioural treatments with the same treatments plus hypnosis. With hypnosis, patients did better on average, particularly strikingly in weight-loss programmes.

A year later, David Allison and Myles Faith recalculated precisely those weight-loss studies. They found transcription and calculation errors. The previously reported mean effect of d = 1.96, an exceptionally large value by usual standards, shrank to 0.26 after correction. Without one questionable study, 0.21 remained, and that value was no longer statistically significant.

This story is not evidence against hypnosis. Allison and Faith re-examined the particularly striking weight-loss finding, not the entire literature on the added benefit of hypnotic procedures. It serves instead as a cautionary tale about how heavily an impressive effect size can depend on a handful of studies and correct arithmetic. In a field full of big promises, it is therefore worth recalculating impressive numbers.

Against this background, we can summarise what the research supports and what it does not.

Positive self-statements that clearly contradict one's self-image have no convincing evidence as a therapeutic tool. For people with low self-esteem, there are experimental indications that they can worsen mood in the short term. More credible are variants that also allow doubts, or sentences that describe a possibility rather than a finished state.

Values-based self-affirmation is a well-researched procedure with small, context-dependent effects. It seems useful above all where people need to take in threatening information, for example before a difficult diagnosis, uncomfortable feedback or an upcoming behaviour change. It takes a few minutes. As a standalone treatment it is not supported.

For procedures in which suggestions are linked to changes that can be directly experienced, the evidence becomes more concrete. Autogenic training and self-hypnosis have clinical evidence of effectiveness, with all their methodological limitations. For experimental pain, hypnotic suggestions are especially well studied, with large differences between individuals. Anyone offering these procedures should take individual responsiveness seriously rather than blaming every failure on lack of practice.

What most people want to know, though, is a different question: do affirmations add anything on top of an effective treatment? Here the studies are simply missing. I am not aware of any convincing study showing a clinically meaningful added benefit over established approaches such as cognitive restructuring, self-compassion exercises or concrete behavioural planning. That is not proof that such a benefit does not exist. But it is a good reason not to promise it.

The sections of this article lead to four practical consequences.

If you are looking for a sentence that is supposed to help, ask whether something you can directly experience can follow it. "I am calm" can hardly be checked in a moment of anxiety. "With the next out-breath it will get a little easier" can be checked, and the check often comes out positive.

If you work with suggestion or relaxation, start with changes that are likely to happen and easy to notice. A first success is worth more than an ambitious formula that is not confirmed.

If you experience such a success, you should be allowed to credit yourself with it. After a successful exercise, the question "What did you just do yourself?" is probably more useful than "See, the method works". That this attribution makes a lasting difference is a hypothesis. The direction of the existing findings speaks in its favour.

And anyone who stands in front of the mirror in the morning does not have to stop. But they can change something. Instead of talking themselves into a finished state, they can recall something that matters to them and that they have already lived. Or they can aim for a small change that can happen by evening.

This brings us back to the beginning. The question from the first section was why a brain should adopt a claim for which it is simultaneously holding counterevidence. This article's answer: a claim does not become valuable because it is repeated often enough. It becomes interesting where a fitting experience follows it and the person can learn something from that experience about their own capacity to act. Evidence of that kind cannot be talked into existence. It can only be experienced.

← Back to the blog · Deutsche Fassung →